DeepPathway: Predicting Pathway Expression from Histopathology Images
Spatial transcriptomics (ST) technologies provide spatially resolved gene expression along with image data, allowing the integrative analysis of complex tissue microenvironments. Despite their potential, the widespread adoption of ST remains limited due to high costs, and methodological challenges in data acquisition. Thus, there have been recent efforts to develop deep learning methods capable of inferring spatial gene expression from the much cheaper and easily available haematoxylin and eosin (H&E) images. These methods demonstrate promising results in reconstructing transcriptomic landscapes within tissue sections. While existing approaches predominantly focus on gene-level predictions, biological processes are often regulated at the pathway level through coordinated activity among functionally related genes. We present DeepPathway, a contrastive learning-based approach trained on ST data to predict pathway expression from H&E-stained sections. We compute input pathway expression by summarizing the expression of constituent genes using established pathway definitions. We evaluate the performance of our method on two prostate cancer datasets and validate our approach on the H&E images acquired from The Cancer Genome Atlas (TCGA) clearly differentiating between normal and tumour tissues. Finally, we apply our method to predict hypoxia signatures using H&Es of brain tumour samples where hypoxia staining with pimonidazole was available as ground truth. Implementation code for DeepPathway is available at https://github.com/aahsan045/DeepPathway.